Full opinion text
McCONNELL, Circuit Judge. Michael Lee Wilson, a death row inmate in the Oklahoma State Penitentiary, appeals the district court’s denial of his petition for a writ of habeas corpus filed pursuant to 28 U.S.C. § 2254. Mr. Wilson was convicted of one count of murder in the first degree and robbery with a dangerous weapon. In the sentencing phase, the jury found three statutory aggravating factors. He was sentenced to death for the first degree murder and to life in prison for the robbery. For the reasons set forth below, we affirm the district court as to all issues other than ineffective assistance of counsel at the mitigation phase; with respect to that issue we remand for an evidentiary hearing. Judge Hartz and Judge Tymkovich join all but Part III of this opinion, which addresses the ineffective assistance of counsel claim. Judge Hartz joins Part III(c) and concurs in the result of Part III. Judge Tymkovich dissents from the holding of Part III. I. Background The factual findings of the Oklahoma Court of Criminal Appeals (“OCCA”) are presumed correct unless rebutted by clear and convincing evidence. 28 U.S.C. § 2254(e)(1). Our account of the crime and the trial is based on the OCCA’s opinion in Wilson v. State, 983 P.2d 448 (Okla.Crim.App.1998) (“Wilson I ”). A. The Crime Michael Lee Wilson, along with Billy Alverson, Darwin Brown, and Richard Harjo, planned to rob the QuikTrip convenience store in Tulsa, Oklahoma, where Mr. Wilson was an employee. The planning occurred for approximately two weeks prior to the crime. At 11:00 p.m. on the night of February 25,1995, Mr. Wilson completed his shift at the QuikTrip and was replaced by Richard Yost. In the early morning hours of February 26, Mr. Wilson, along with Alverson, Brown, and Har-jo, entered the QuikTrip. The subsequent events were captured on the store’s surveillance tape. Wilson I, 983 P.2d at 455. The four found Yost cleaning the windows on the coolers and surrounded him. When Yost tried to walk away, the four attacked him and dragged him into the back room. What occurred in that room is not visible on the surveillance tape, but the tape captured noises coming from there. At some point, Yost was handcuffed. After helping to drag Yost to the back room, Alverson exited, picked up some items that had been knocked from the shelves during the struggle, and kept watch for customers. Soon after, Harjo also left the back room and the two walked out of the store together. As they exited, Yost yelled and screamed for help, believing that a customer had entered the store. When Alverson and Harjo returned, Harjo was carrying a black aluminum baseball bat. Both returned to the back room. Id. Sounds of the bat striking Yost are audible on the surveillance tape. In addition to striking his head, trial evidence suggested that the baseball bat struck the handcuffs on Yost’s wrists as he held his hands above his head in an attempt to block the blows. These blows eventually caused Yost’s death. Id. During the attack, Mr. Wilson exited the back room, examined his hands, put on a QuikTrip jacket, and went behind the counter. He greeted customers as they entered and completed sales, attempting to remove the safe below the counter between customers. He eventually succeeded. He also took money from the cash drawer and from the currency change machine, and removed the video from the surveillance camera recorder. Using a dolly from the store, the defendants loaded the safes and money into Mr. Wilson’s car. Id. A customer discovered Yost’s dead body soon after the defendants left the Quik-Trip. It was in a gruesome condition. Yost’s ankles had been taped together with duct tape. There was a handcuff near his body; a piece was later discovered embedded into his head. His body was lying in a pool of spilled milk, beer, and blood. After speaking to witnesses, the police discovered that Mr. Wilson had been in the store between 4:00 a.m. and 6:00 a.m. When he failed to come to work during his scheduled 3:00 p.m. shift that day, Officer Wayne Allen set up surveillance at Mr. Wilson’s home. Around 5:00 p.m., a gray vehicle pulled up to Mr. Wilson’s house. Mr. Wilson got out of the car, picked up a shovel and waived it in the air; soon after, he returned to the vehicle and the car drove away. Officer Allen stopped the car and arrested Mr. Wilson along with Alver-son, Harjo, and Brown, who were also inside. The police discovered large sums of money on all of the defendants except for Mr. Wilson. Id. Following the arrest, Detective Charles Folks questioned Mr. Wilson. Part of the questioning was recorded, but another component was not. In an unrecorded component, Mr. Wilson indicated that he knew Yost would be killed, and that the four had planned the robbery for approximately two weeks. Mr. Wilson was to act as the sales clerk after Yost was “taken care of.” Id. When the police searched Alverson’s home, they discovered the drop safe, the dolly, QuikTrip glass cleaner, money tubes, and the surveillance video tape. Nothing was discovered at Mr. Wilson’s home. However, on February 27, Mr. Wilson’s mother, Patricia Taylor, requested that the police come to her house. When they arrived, they found a baseball bat, a bloody QuikTrip jacket with Mr. Yost’s name on it, Mr. Wilson’s Nike jacket, (which matched the one he wore during the robbery), and the other handcuff, all placed on the front porch. Id. at 455-56. B. Judicial Proceedings Mr. Wilson was tried, along with Darwin Brown, before a jury in the Tulsa County District Court. The court used a dual jury procedure, where one trial was conducted in front of two juries. One jury was assigned to adjudicate the charges against Mr. Brown while a separate jury was assigned to Mr. Wilson’s case. Wilson I, 983 P.2d at 456. The jury convicted Mr. Wilson of first degree murder and robbery with a dangerous weapon. At the sentencing phase, the Bill of Particulars charged that Mr. Wilson should be punished by death due to the existence of three aggravating circumstances: (1) the murder was especially heinous, atrocious, or cruel; (2) the murder was committed for the purpose of avoiding or preventing a lawful arrest or prosecution; and (3) it was probable that the defendant would commit criminal acts of violence that would constitute a continuing threat to society. Okla. Stat. Ann. tit. 21, § 701.12(4),(5), & (7). The jury found the existence of all three aggravating circumstances and recommended that Mr. Wilson be sentenced to death. The trial court agreed, and sentenced Mr. Wilson to death. The OCCA affirmed both Mr. Wilson’s conviction for murder in the first degree and his death sentence. It reversed his conviction and sentence for robbery with a dangerous weapon; because Mr. Wilson was convicted of felony murder, under Oklahoma law, he could not be convicted of the underlying felony. Wilson I, 983 P.2d at 463. The United States Supreme Court denied Mr. Wilson’s petition for writ of certiorari on October 4, 1999. Wilson v. Oklahoma, 528 U.S. 904, 120 S.Ct. 244, 145 L.Ed.2d 205 (1999). The OCCA denied post-conviction relief on November 15, 1999. R. Vol. Ill Box 2, “Opinion Denying Posb-Conviction Relief’ (“Wilson II”). Mr. Wilson filed a petition for a writ of habeas corpus pursuant to 28 U.S.C. § 2254 in the District Court for the Northern District of Oklahoma on February 16, 2000. Wilson v. Sirmons, No. 00-CV-147, 2006 WL 2289777 (N.D.Okla. Aug. 8, 2006) (‘Wilson III”). The district court denied the petition on August 8, 2006, but granted a certificate of appealability (“COA”) pursuant to 28 U.S.C. § 2253(c)(2) on fourteen grounds: (1) the use of the dual jury procedure; (2) the introduction of DNA evidence without a Daubert hearing; (3) the introduction of evidence in violation of the Oklahoma discovery code; (4) the failure to give jury instructions on a lesser included offense; (5) the introduction at trial of Mr. Wilson’s statements obtained without Miranda warnings; (6) the admission of prejudicial hearsay; (7) the sufficiency of the evidence to support the heinous, atrocious, or cruel aggravator and the constitutionality of the heinous, atrocious, or cruel aggravator; (8) the introduction of irrelevant and prejudicial evidence; (9) prosecutorial misconduct; (10) the improper sequence of voir dire questioning; (11) the improper use of victim impact evidence; (12) ineffective assistance of counsel; (13) application of the continuing threat aggravator and the constitutionality of the continuing threat ag-gravator; and (14) the cumulative impact of the errors on Mr. Wilson’s rights. II. Standard of Review Mr. Wilson filed his habeas corpus petition after the enactment of the Antiterrorism and Effective Death Penalty Act of 1996 (“AEDPA”). Therefore, AEDPA’s standards apply to all claims adjudicated on the merits in state court proceedings. Under AEDPA, a writ of habeas corpus will not be granted unless the state court’s adjudication of the claim (1) resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established federal law, as determined by the Supreme Court of the United States; or (2) resulted in a decision that was based on an unreasonable determination of the facts in light of the evidence presented in the state court proceeding. 28 U.S.C. § 2254(d)(l)-(2); Turrentine v. Mullin, 390 F.3d 1181, 1188 (10th Cir.2004). We must presume that the state court’s factual findings are correct unless the petitioner rebuts the presumption by clear and convincing evidence. 28 U.S.C. § 2254(e)(1); Smith v. Mullin, 379 F.3d 919, 924-25 (10th Cir.2004) (quoting Smallwood v. Gibson, 191 F.3d 1257, 1264-65 (10th Cir.1999)). The district court’s legal analysis is reviewed de novo and any factual findings are reviewed for clear error. Turrentine, 390 F.3d at 1189; Smallwood, 191 F.3d at 1264-65. A state court decision is “contrary” to clearly established law “if the state court applies a rule different from the governing law set forth in [Supreme Court] cases, or if it decides a case differently than [the Supreme Court has] done on a set of materially indistinguishable facts.” Bell v. Cone, 535 U.S. 685, 694, 122 S.Ct. 1843, 152 L.Ed.2d 914 (2002). A state court decision is an “unreasonable application” of clearly established law when the state court “identifies the correct governing legal principle from [the Supreme Court’s] decisions but unreasonably applies that principle to the facts of petitioner’s case.” Wiggins v. Smith, 539 U.S. 510, 520, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003) (citing Williams v. Taylor, 529 U.S. 362, 413, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000)). “[T]he state court’s decision must have been more than incorrect or erroneous ... [it] must have been ‘objectively unreasonable.’ ” Id. at 520-21, 123 S.Ct. 2527 (citing Williams, 529 U.S. at 409, 120 S.Ct. 1495). If we find that the state court erred, we still must determine whether the error is a structural defect “in the constitution of the trial mechanism, which def[ies] analysis by ‘harmless-error’ standards.” Arizona v. Fulminante, 499 U.S. 279, 309, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991). If it is not a structural error, then we apply the harmless-error standard articulated in Brecht v. Abrahamson, 507 U.S. 619, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993), and in O’Neal v. McAninch, 513 U.S. 432, 115 S.Ct. 992, 130 L.Ed.2d 947 (1995). Under the harmless-error test, relief is only proper if the error “had substantial and injurious effect or influence in determining the jury’s ver-diet.” Brecht, 507 U.S. at 623, 113 S.Ct. 1710 (citing Kotteakos v. United States, 328 U.S. 750, 776, 66 S.Ct. 1239, 90 L.Ed. 1557 (1946)); see also Fry v. Pliler, — U.S. -, 127 S.Ct. 2321, 2327-28, 168 L.Ed.2d 16 (2007). A ‘“substantial and injurious effect’ exists when the court finds itself in ‘grave doubt’ about the effect of the error on the jury’s verdict.” Bland v. Sirmons, 459 F.3d 999, 1009 (10th Cir.2006) (citing O’Neal, 513 U.S. at 436, 115 S.Ct. 992). There is “grave doubt” when the issue of harmlessness “is so evenly balanced that [the court] feels [itself] in virtual equipoise as to the harmlessness of the error.” O’Neal, 513 U.S. at 435, 115 S.Ct. 992. If the state court did not decide the claim on the merits, the stringent principles of deference under 28 U.S.C. § 2254 are inapplicable. Mitchell v. Gibson, 262 F.3d 1036, 1045 (10th Cir.2001). Additionally, if the district court’s factual findings are entirely dependent on the state court record, our review of those findings is de novo. Smallwood, 191 F.3d at 1265 n. 1. III. Ineffective Assistance of Counsel Petitioner’s most persuasive claim is that he was deprived of effective assistance of counsel because of his trial attorney’s failure “to adequately prepare his mental health expert, Dr. Eugene Reynolds, to testify in second stage or even make use of all of the mitigating information about Petitioner’s mental state that Dr. Reynolds could have provided to the jury.” Aplt. Br. 71. Mr. Wilson argues that he is entitled to an evidentiary hearing on this matter, which he has yet to receive. If not barred by AEDPA, a defendant is entitled to an evidentiary hearing “so long as his allegations, if true and not contravened by the existing factual record, would entitle him to habeas relief.” Miller v. Champion, 161 F.3d 1249, 1253 (10th Cir.1998). In recent years, the Supreme Court and this Court have placed increasing emphasis on the obligation of defense counsel in capital cases to develop and present mitigating evidence in the penalty phase of the trial, often on the basis of family upbringing and mental health. This is a closer case than some, because defense counsel did hire an appropriate expert, provide some background information, and present some of the expert’s findings to the jury. Cf. Anderson v. Sirmons, 476 F.3d 1131 (10th Cir.2007) (reversing denial of habeas relief where defense counsel utterly failed to present mitigating evidence based on family history and mental health). The Supreme Court, however, has made clear that the investigation and presentation of some mitigating evidence is not sufficient to meet the constitutional standard, if counsel fails to investigate reasonably available sources or neglects to present mitigating evidence without a strong strategic reason. Wiggins v. Smith, 539 U.S. 510, 527, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003); American Bar Association Guidelines for the Appointment and Performance of Counsel in Death Penalty Cases 11.8.6 (1989) [hereinafter “ABA Guidelines”]. As we said in Romano v. Gibson, 239 F.3d 1156, 1180 (10th Cir.2001): “The sentencing stage is the most critical phase of a death penalty case. Any competent counsel knows the importance of thoroughly investigating and presenting mitigating evidence.” Because Mr. Wilson’s allegations, if true and fully developed, would entitle him to relief, we reverse the district court’s denial of an evidentiary hearing on this claim, and remand to the district court. A. Factual Background 1. Pre-trial Preparation Although counsel was appointed to defend Mr. Wilson some two years before trial, he waited until just three weeks before trial started before he contacted and hired a mental health expert to assist in mitigation. This expert was Dr. Eugene Reynolds, a clinical psychologist. No explanation for this delay appears in the record. Dr. Reynolds was able to visit Mr. Wilson three times prior to the sentencing phase, but only during the first visit did the two have any privacy. Pet. Addendum 2, at ¶ 2; Tr. trans. 2/19/97, at 53-54. The second two visits were conducted in a cubicle in the hallway, with police officers and inmates walking by within earshot. Pet Addendum 2, at ¶ 2. Trial counsel provided Dr. Reynolds with the following materials: Hillcrest Medical Center Records, Scholastic School Records, Children’s Medical Center Records, and statements by five individuals, three of whom testified at trial. None of those statements came from family members. During his interviews with Mr. Wilson, Dr. Reynolds administered several psychological tests from which he formulated three major conclusions: (1) that Mr. Wilson had an IQ score of 126; (2) that there was no evidence of neurological or organic brain damage; and (8) that Mr. Wilson suffered from generalized anxiety disorder, bipolar disorder (severe without psychotic features), and post-traumatic stress disorder (PTSD). The testing also indicated paranoid personality disorder and narcissistic personality disorder with passive-aggressive and schizotypal personality features. One test suggested Mr. Wilson suffered from schizophrenia; however, that test was invalid. According to Dr. Reynolds, “additional testing” and “further collateral data” were required to support this diagnosis, but “[ujnfortunately, there wasn’t enough time to obtain this information before the trial.” Pet. Addendum 2, at ¶ 4. Twelve days before his testimony, which was four days after jury selection had already begun, Dr. Reynolds made his results available to trial counsel, though it is unclear in what form. We do know, however, that counsel did not meet with Dr. Reynolds to discuss these results until the day before the sentencing phase began — two days before Dr. Reynolds’ testimony. The only family member counsel made available to Dr. Reynolds was Ms. Patricia Taylor, Mr. Wilson’s mother, and Dr. Reynolds spoke to her only after he completed his testing. At no point did counsel himself interview Ms. Taylor about Mr. Wilson’s life. Neither counsel nor Dr. Reynolds spoke to any other family member. Mr. Wilson has a sister and a brother, as well as a girlfriend, with whom he has a child. 2. Mitigation Phase at Trial During the mitigation phase, counsel called six witnesses to testify for Mr. Wilson. Two individuals knew Mr. Wilson through church, but they could provide only limited observations about Mr. Wilson, including that he was “mannerable,” “respectful,” and “intelligent.” Tr. trans. 2/19/97, at 13, 19, 22. Two of Mr. Wilson’s former teachers also testified. Because they had not seen him in approximately five to six years, they also provided only limited insight, describing Mr. Wilson as “respectful,” “fun-loving,” and a “very good student.” Id. at 35, 38. Counsel’s most important witness was Dr. Reynolds. On direct examination, counsel asked Dr. Reynolds generally about the tests administered to Mr. Wilson. He asked Dr. Reynolds about Mr. Wilson’s high IQ of approximately 126, placing Mr. Wilson in the “superior range of intelligence” category, which, Dr. Reynolds stated, indicated that Mr. Wilson could “do something with himself.” Tr. trans. 2/19/97, at 55-56, 63. Counsel asked Dr. Reynolds only a few additional questions about the results of his testing. In response, Dr. Reynolds testified that Mr. Wilson experienced a “severe mental disorder with many of the personality scales elevated. That would suggest that he has a severe personality disturbance.” Id. at 57. Despite the fact that Dr. Reynolds had other diagnoses to give, counsel asked him no further questions about the specific results and conclusions from the psychiatric testing. Counsel asked briefly about Mr. Wilson’s social history; Dr. Reynolds described in a few sentences Mr. Wilson’s father as someone who was active in “drugs and alcohol, and pretty much ... not involved in Michael’s life.” Id. at 59. Dr. Reynolds focused primarily on the “two pictures” of Mike. “On the one hand, you have the picture of the Sunday school-going child. On the other hand, you have the picture of the gang and the uninvolved father, who did not set a particularly good role model.” Id. at 60. At no point did counsel elicit Dr. Reynolds’ more concrete, scientifically rooted diagnoses, including the PTSD, bipolar disorder, generalized anxiety disorder, and schizotypal personality features. The entirety of Dr. Reynolds’ description of Mr. Wilson’s psychological state is no more than a page of the sentencing transcript. What occurred on cross-examination was a train wreck for Mr. Wilson. The prosecutor asked Dr. Reynolds: Q: [A]re there psychopathic criminals who have superior intelligence? A: Yes. Q: The sharp contrast [Mr. Wilson] exhibits ... aren’t those classic designs of a psychopath? ‘Yes’ or ‘no’? A: It can be. Q: And aren’t psychopaths the most likely to re-offend, based on the studies? A: Yes. Tr. trans. 2/19/97, at 65. Later in the cross examination, the prosecutor continued: Q: [A]ren’t superficial charm and good intelligence, coupled with cunning and manipulative lack of imp[ul]sivity behavior characteristics of a psychopath? A: Yes, they are. Q: And that’s what Mr. Wilson has, isn’t it? A: Some of those characteristics, he has. Tr. trans. 2/19/97, at 76. In his closing argument, the prosecutor used this testimony to again call Mr. Wilson a “psychopathic killer based on the evidence.” Tr. trans. 2/20/97, at 46. Finally, Mr. Wilson’s mother testified at trial. She had not talked with defense counsel at any point prior to her testimony. She spoke briefly about Mr. Wilson’s father and discussed Mr. Wilson’s involvement at church. That Ms. Taylor had more to say was apparent from her statement, after defense counsel finished his questioning, that she “did want to say something else, if I’m allowed.” Tr. trans. 2/19/97, at 103. Because defense counsel had rested, the court could not permit her to do so. 3. Postr-Conviction Investigation By Appellate Counsel On direct appeal, new counsel was appointed to represent Mr. Wilson. Appellate counsel provided Dr. Reynolds with additional information, including Tulsa County Gang Intervention Team Records, hospital records, Tulsa public school records, and, most significantly, affidavits from Ms. Taylor, James Leon Wilson (Mr. Wilson’s brother), Staci Faenze (Mr. Wilson’s sister), and Tonya Holt (Mr. Wilson’s former girlfriend and the mother of his child). With this information as background material, Dr. Reynolds performed' a second set of tests. These tests supported a diagnosis of schizophrenia, paranoid type. Typical symptoms include “delusions, hallucinations, disorganized speech, disorganized or catatonic behavior and negative” manifestations. Pet. Addendum 2, at ¶7, 11. Testing also revealed that Mr. Wilson had a “severe psychological disturbance with the possibility of delusions or hallucinations.” Id. at ¶ 7. Dr. Reynolds reported that Mr. Wilson believed “evil spirits” possessed him at times, and that it was “possible [Mr. Wilson] could have been delusional at the time of the crime.” Id. at ¶ 10. Interviews with the family members revealed that Mr. Wilson suffered from depression, concentration problems, and delusions, that he heard voices, and that he frequently experienced memory lapses. At one point, when his girlfriend informed Mr. Wilson she heard voices, Mr. Wilson responded “I’ve heard them too, I hear voices and its OK. You just have to fight them, you just have to pray them away and they will go away.” Pet. Addendum 4, at ¶ 11. All of the family members, as well as his girlfriend, vividly described the violent nightmares from which Mr. Wilson suffered throughout his life, during which he would kick, punch and shout all night. Mr. Wilson often experienced severe headaches that lasted for hours and sometimes days. The affidavits also highlighted several experiences during Mr. Wilson’s youth that may have led to his emotional and mental problems, including his relationship with his father. Although there was testimony at trial that Mr. Wilson’s father was uninvolved in his life, the affidavits explained the effect that this absence, and his father’s constant drug use in and out of the home, had on Mr. Wilson. Mr. Wilson’s older brother, with whom Mr. Wilson was very close, sold dope to support the family and eventually became hooked on crack cocaine. He, like his father, was in and out of prison. The brother was heavily involved in a gang from the time Mr. Wilson was young, and Mr. Wilson in turn grew up surrounded by and involved with the same individuals. Mr. Wilson’s brother stated that gang members fired shots at Mr. Wilson “at least once every week.” Pet. Addendum 5, at ¶ 4. Dr. Reynolds stated in his affidavit that “[t]hese affidavits and additional testing helped me reach a more accurate diagnosis since this information was not previously provided during my first evaluation.” Pet. Addendum 2, at ¶ 7. He concluded: My testimony could have been improved upon enormously had I been provided with the additional information provided to me by the Appellate Defense Council [sic]. This information provides the history of Micheál [sic] experiencing delusions and hallucinations, and other behaviors which supports the diagnosis of schizophrenia, paranoid type. Knowing this may have helped the jury better understand Micheal’s [sic] emotional illness and how he could have participated in the crime. Id., ¶ 15. A The OCCA Decision On direct appeal, the OCCA rejected Petitioner’s argument that trial counsel’s representation with respect to mental health mitigation was constitutionally deficient. As explained more fully below, the OCCA made no reference to the post-conviction investigation or affidavits, relying entirely on the record at trial. After briefly summarizing Dr. Reynolds’ preparation, the court concluded: Reynolds testified that Wilson had a severe personality disturbance. Reynolds explained that Wilson had some unusual, bizarre types of thinking that would suggest that he is not in touch with reality at times. Reynolds [sic] testimony indicated that Wilson committed this crime as an intelligent but immature person, and that, because of his family support and his intelligence, he had the capability of being rehabilitated. The mere fact more evidence could have been presented is not, in itself, sufficient to show counsel was deficient. Reynold’s [sic] testimony was credible and well developed. We find Appellant has failed to carry his burden to show either deficient performance by counsel, or prejudice from the omission of this specific evidence. Wilson I, 983 P.2d at 472 (citation omitted). The court made no finding that any of the alleged deficiencies in trial counsel’s performance were a product of strategic judgment. It denied Petitioner’s motion for an evidentiary hearing. 5. The District Court Decision The district court denied the petition based on the OCCA’s analysis, which the court found was not an unreasonable application of Supreme Court precedent. Wilson III, 2006 WL 2289777, at *43. According to the district court: A careful reading of the trial transcript confirms that Petitioner’s trial counsel questioned Dr. Reynolds thoroughly during his second stage testimony. This Court finds nothing deficient in the performance of trial counsel. Accordingly, the Court finds that the OCCA’s rejection of this claim on direct appeal was not an unreasonable application of the legal principle announced by the Supreme Court in Strickland to the facts of Petitioner’s case. Petitioner has failed to satisfy the § 2254(d) standard on this portion of his ineffective assistance of counsel claim. Id. The district court made no finding on prejudice, and also denied Mr. Wilson’s request for an evidentiary hearing. B. The State’s Nonresponsive Brief In its brief in this Court, the State offers almost no defense of counsel’s performance. The entirety of its argument is found in this short paragraph: Trial counsel hired Dr. Reynolds to provide a complete evaluation of Appellant’s mental health. In addition, counsel provided information to Dr. Reynolds and made the Appellant and others available to assist in the diagnosis. Trial counsel did not provide deficient performance. See Trice v. Ward, 196 F.3d 1151, 116[sic] (10th Cir.1999). Resp. Br. 80. Significantly, the State offers no reason we should regard the affidavits based on post-conviction investigation as either procedurally or substantively deficient, and no argument that any of the alleged deficiencies in trial counsel’s performance were the product of strategic judgment. If the entirety of counsel’s obligation to present mitigation evidence based on mental health were to hire an expert (no matter when), to provide him some “information,” and to arrange for access to the defendant and one family member, this would be responsive. It is unresponsive when the defendant has introduced specific evidence indicating that counsel hired the expert so late in the process that he was unable to complete necessary mental health evaluations, that counsel failed to gather or provide readily available relevant information that would have affected the diagnosis, and that counsel failed to present the expert’s actual diagnoses to the jury. It bears mention, therefore, that much of the dissenting opinion’s analysis is based on arguments not developed, or even hinted at, by the State’s brief. C. The Need for De Novo Review As already.noted, the district court denied Petitioner’s request for habeas relief, or even an evidentiary hearing, on the basis of deference to the OCCA’s decision rejecting Petitioner’s arguments on direct appeal, as ordinarily required by 28 U.S.C. § 2254(d). Wilson III, 2006 WL 2289777, at *43. The State similarly rests on this deferential standard of review in arguing for affirmance. Resp. Br. 78. We cannot follow this path. While we review a district court’s denial of an evidentiary hearing for abuse of discretion, see Coronado v. Ward, 517 F.3d 1212, 1217 (10th Cir.2008), it is well established in this Circuit that when a state court’s disposition of a mixed question of law and fact, including a claim of ineffective assistance, is based on an incomplete factual record, through no fault of the defendant, and the complete factual record has since been developed and is before this Court, we apply de novo review to our evaluation of the underlying claim. Bryan v. Mullin, 335 F.3d 1207, 1215 (10th Cir.2003) (en banc); see Miller, 161 F.3d at 1254 (when “the state court [does] not hold an evidentiary hearing” on non-record evidence proffered on direct appeal, the Court is in “the same position to evaluate the factual record as [the state court] was.”). When the state court makes factual findings on the basis of an incomplete factual record in such a case, “we need not afford those findings any deference.” Miller, 161 F.3d at 1254. Of course, to obtain an evidentiary hearing on habeas, Mr. Wilson has the burden of first showing that he “diligently sought to develop the factual basis underlying his habeas petition, but a state court prevented him from doing so,” or that his evidentiary proffer falls within an exception to 28 U.S.C. § 2254(e)(2)’s ban on the admission of new evidence. Miller, 161 F.3d at 1253. But the State does not argue that Mr. Wilson failed to satisfy this requirement, and we find that he acted diligently before the OCCA. To be clear, this does not mean that we apply de novo review every time the state court declines to hold a hearing on a defendant’s evidentiary proffer. Had the state court evaluated the non-record evidence in its denial of Mr. Wilson’s Strickland claim and his request for an evidentiary hearing, we would apply AEDPA’s deferential standard. See, e.g., Welch v. Sirmons, 451 F.3d 675, 704, 708-09 (10th Cir.2006) (applying AEDPA deference when the OCCA referred to the proffered affidavits in the course of denying an ineffective assistance claim); Bland v. Sirmons, 459 F.3d 999, 1030 (10th Cir.2006) (the OCCA “[e]xam-in[ed] the affidavits that would be proffered at such a[n] [evidentiary] hearing.”). In these cases, the state court examined the claim on the merits, including the proffered non-record evidence, but decided that even if that new evidence were fully developed, the defendants could not meet their burdens under Strickland. As we explain more fully below, the OCCA in this case, by contrast, made clear that it was relying solely on the trial record, and not the non-record evidence, when it denied the claim and the evidentiary hearing. Wilson I, 983 P.2d at 472 & n. 8 (“[a] review of the trial record shows trial counsel did put forth mental health expert.... Accordingly, we further find that Wilson’s application for an evidentiary hearing on this claim should be denied.”). Because Mr. Wilson did not receive consideration of his non-record claim, Bryan, 335 F.3d at 1215, and Miller, 161 F.3d at 1254, govern the standard of review in this case. Because of the importance of standard of review to this case, we now set forward in detail the reasons we conclude both that Mr. Wilson was diligent in pursuing his non-record claims and that the OCCA did not consider his non-record evidence in the course of affirming the sentence and denying his application for an evidentiary hearing. Assigned new counsel on direct appeal to the OCCA, Petitioner asserted that his trial counsel was ineffective during the sentencing phase of trial because he failed to properly investigate Mr. Wilson’s mental health background and to adequately prepare the expert witness. Petitioner also claimed that defense counsel did not properly present those mental health diagnoses that the expert did make prior to trial. Petitioner proffered five affidavits in support of these claims. Three of these affidavits were from family members and one was from Mr. Wilson’s former girlfriend, who is also the mother of his child. All described different mental health problems Mr. Wilson had, along with other struggles he experienced throughout his youth. The fifth was from the trial expert, Dr. Eugene Reynolds. In his affidavit, Dr. Reynolds described the diagnoses he reached after receiving the additional affidavits from Mr. Wilson’s family, provided by appellate counsel. He also outlined the diagnoses he made prior to the sentencing phase but which were never presented to the jury, and recounted his interactions with the defendant. To develop his ineffective assistance claim and bring this non-record evidence into the record, Mr. Wilson requested an evidentiary hearing under Oklahoma Court of Criminal Appeals’ Rule 3.11(B)(3)(b). Okla. Stat. tit. 22, ch. 18, App. Rule 3.11(B)(3)(b). This was sufficient to meet the “diligence” requirement of 28 U.S.C. § 2254(e)(2), and the State does not contend otherwise. The question, then, is whether the state court considered this evidence when it rendered its decision. If so, its decision is entitled to deference; if not, we must make our own de novo evaluation. Oklahoma Appellate Rule 3.11(B)(3)(b) allows a defendant, on direct appeal, to offer non-record evidence in support of an ineffective assistance of trial counsel claim. If the court finds, “by clear and convincing evidence there is a strong possibility trial counsel was ineffective for failing to utilize or identify the complained-of evidence,” the OCCA will remand to the trial court for an evidentiary hearing based on the claims raised in the application. Okla. Stat. tit. 22, ch. 18, App. Rule 3.11(B)(3)(b); Dewberry v. State, 954 P.2d 774, 775 (Okla.Crim.App.1998). Following the evidentia-ry hearing, the trial court makes written findings of fact and conclusions of law. “It is the record from this evidentiary hearing which ... supplements the trial court record on appeal.” Dewberry, 954 P.2d at 776. Any affidavits or other evidence supplied in support of the evidentiary hearing are not considered part of the record on which the OCCA bases its Strickland ruling unless they are properly introduced at the evidentiary hearing. Id. (“The record on appeal must remain as only that which has been presented through the trial court.”) (emphasis added). If it does not remand for an evidentiary hearing or consider the non-record evidence in the course of denying the application for an evidentia-ry hearing, the OCCA’s ultimate ruling on the Strickland issue cannot be based on the non-record information proffered in support of the claim. Judge Tymkovich argues that “[bjecause the standard for obtaining an evidentiary hearing under Rule 3.11 is lower than the standard set forth in Strickland — petitioner need only show a ‘strong possibility’ of ineffective assistance — when the OCCA denies an evidentiary hearing under Rule 3.11, it necessarily makes a merits determination petitioner cannot meet the substantive Strickland, standard.” Diss. Op. 1129. We cannot accept this interpretation. Although Rule 3.11 uses a lower substantive standard (“strong possibility”) it erects a much higher evidentiary hurdle for meeting that standard: to obtain an evidentiary hearing under Rule 3.11, the movant must provide “clear and convincing evidence” of this “strong possibility.” The federal standard does not impose this “clear and convincing evidence” hurdle. To receive an evidentiary hearing on a Strickland claim, a petitioner need show only that “the allegations, if true and not contravened by the existing factual record, would entitle him to habeas relief.” Miller, 161 F.3d at 1249. Although the interplay of these two standards — one more demanding, one less demanding, than the federal — is not clear, we cannot conclude that when the state court denies an evidentiary hearing under Rule 3.11 it has necessarily decided that the federal standard was not satisfied. The OCCA did not remand for an evi-dentiary hearing, Wilson I, 983 P.2d at 472 n. 8. Nor did it consider the affidavits in the course of denying Mr. Wilson’s application for an evidentiary hearing. The entirety of the OCCA’s analysis of the Strickland issue was based on the court’s review of the trial record. Id. at 472 (“A review of the trial record shows that counsel did put forth a mental health expert....”). The court’s sole reference to the affidavits was to note that “Wilson has filed, contemporaneously with this issue, an application for an evidentiary hearing ... in an attempt to supplement the record ...,” id., and then. to deny the application. This was consistent with the State’s litigation position on appeal. The State’s appellate brief to the OCCA stated that “the information contained with the affidavits ... are [sic] not being referred to in this brief as such affidavits are not part of the record on appeal.” Brief of Respondent-Ap-pellee at 70 n. 40, Wilson I, 983 P.2d 448 (Okla.Crim.App.1998). At oral argument before this Court, counsel for the State conceded that the affidavits would not have been part of the OCCA’s Strickland determination. See Bryan, 335 F.3d at 1215 (applying de novo review where the defendant sought an evidentiary hearing before the OCCA and attached affidavits, but the OCCA decided the case with only a minor reference to the new evidence); accord Miller, 161 F.3d at 1254. Had the district court held an evidentia-ry hearing or otherwise made any factual findings based on the affidavits, we would accept those findings unless they were clearly erroneous. Bryan, 335 F.3d at 1216. However, the district court relied only on a “careful reading of the trial transcript” in denying Mr. Wilson’s ineffective assistance claim. . Wilson III, 2006 WL 2289777, at *43. We therefore have no choice but to review both legal and factual findings de novo, in light of the affidavits filed and the allegations made by the defendant but not considered by the OCCA or the district court. For purposes of determining whether Mr. Wilson is entitled to an evidentiary hearing on his Strickland claim, we treat these allegations as true. Miller, 161 F.3d at 1258. The dissent argues that Schriro v. Landrigan, — U.S.-, 127 S.Ct. 1933, 167 L.Ed.2d 836 (2007), overruled our prior precedent requiring de novo review' of timely-filed habeas claims involving non-record evidence not considered by the state court. We do not read it that way. In Schriro, the Supreme Court stated: “Because the deferential standards prescribed by § 2254 control whether to grant habeas relief, a federal court must take into account those standards in deciding whether an evidentiary hearing is appropriate.” Id. at 1940. It is this statement that Judge Tymkovich thinks overrules our prior precedent in Miller and Bryan. See Diss. Op. 1128. Our precedent, however, does consider (or “take into account”) AEDPA’s mandate. Under § 2254(d), AEDPA’s standard of deference applies only to claims that have been “adjudicated on the merits in State court proceedings.” When a state court has not examined the non-record evidence, it has reached no conclusion “on the merits.” See Hoffman v. Arave, 236 F.3d 523, 536 (9th Cir.2001) (“Because the state court denied Hoffman’s ineffective assistance of counsel claims without holding a hearing ... those claims have never been litigated on the merits.”). Schriro does not undermine this analysis. In that case, the defendant — first in his application for state post-conviction relief and later on federal habeas — sought an evidentiary hearing to show that his trial counsel had failed to investigate certain mitigating evidence. The state court declined- to hold an evidentiary hearing on the ground that the defendant had instructed trial counsel not to present any mitigating evidence at the sentencing hearing, rendering any failure to investigate nonprejudicial. The federal district court likewise refused to grant him an evidentia-ry hearing, but the Court of Appeals reversed, finding that the defendant’s instructions referred only to a narrow subset of the potential mitigating evidence. Schriro, 127 S.Ct. at 1938-39. The Supreme Court in turn reversed, finding that the trial record “establish[ed] that the Arizona postconviction court’s determination of the facts was reasonable.” Id. at 1941. It held that the federal courts were bound under AEDPA to defer to the state court’s reasonable finding that the defendant had instructed counsel not to present mitigating evidence, and that this rendered any factual questions regarding counsel’s investigation irrelevant: “If [the defendant] issued such an instruction, counsel’s failure to investigate further could not have been prejudicial under Strickland.” Id. Schriro thus stands for the proposition that if a state court considers non-record evidence and declines to hold an evidentiary hearing — for example, because it finds that the record flatly refutes the evidentiary proffer or renders it irrelevant — the federal court must defer to that determination unless it is “based on an unreasonable determination of the facts” in light of the record. 28 U.S.C. § 2254(d)(2). It was therefore error in Schriro for the Court of Appeals to make its own independent judgment of the facts. But when the state court makes no reference to the new, non-record allegations, either because it simply ignores the information or because it employs a procedural bar against incorporating non-record evidence higher than that permitted by the Constitution, AEDPA deference is not due to the state’s decision not to hold the evi-dentiary hearing, or to any factual determinations made without reference to the proffered evidence. See Bryan, 335 F.3d at 1215-16. We therefore do not regard Schriro as having overruled our prior precedent in Miller and Bryan. We continue to be bound by that precedent. See Dubuc v. Johnson, 314 F.3d 1205, 1209 (10th Cir.2003). Finally, while “federalism, comity, and finality,” Diss. Op. 1126, are undoubtedly important values, the importance of these values is reduced when a claim has never been considered on the merits. Most Strickland claims are based on evidence gathered after the initial trial, which perforce is not part of the original record. A petitioner who has diligently presented such a claim in a timely fashion is entitled to have a court perform a de novo review of his evidence of ineffective assistance. If the state court does not perform this review but instead confínes its review to the original trial record, and the federal court defers to its judgment anyway, that de novo review will never be performed. D. Specific Claims of Defective Performance By Trial Counsel Mr. Wilson argues that trial counsel was ineffective because of his poor investigation in preparation for the sentencing phase and his failure to put on relevant mitigating evidence at trial. “To establish ineffective assistance of counsel, a petitioner must prove that counsel’s deficient performance was constitutionally deficient and that counsel’s deficient performance prejudiced the defense, depriving the petitioner of a fair trial with a reliable result.” Boyd v. Ward, 179 F.3d 904, 913 (10th Cir.1999) (citing Strickland v. Washington, 466 U.S. 668, 687, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984)). Counsel’s performance must be “completely unreasonable” to be constitutionally ineffective, not “merely wrong.” Hoxsie v. Kerby, 108 F.3d 1239, 1246 (10th Cir.1997). When the issue is the adequacy of counsel’s investigation for the sentencing phase of a capital trial,, “hindsight is discounted by pegging adequacy to ‘counsel’s perspective at the time’ investigative decisions are made.” Rompilla v. Beard, 545 U.S. 374, 381, 125 S.Ct. 2456, 162 L.Ed.2d 360 (2005) (citing Strickland, 466 U.S. at 689, 104 S.Ct. 2052). To assess the thoroughness of counsel’s investigation and counsel’s overall performance, the Court must conduct an objective review measured for “reasonableness under prevailing professional norms.” Wiggins v. Smith, 539 U.S. 510, 523, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003) (citing Strickland, 466 U.S. at 688, 104 S.Ct. 2052). We are “highly deferential” to counsel’s decision, and a petitioner must “overcome the presumption that counsel’s conduct was not constitutionally defective.” Wallace, 191 F.3d at 1247. Our analysis today is guided by the Supreme Court’s recent jurisprudence emphasizing the importance of thorough investigation — in particular, of mental health evidence — in preparation for the sentencing phase of a capital trial. While initially, the Supreme Court applied Strickland rather narrowly, see, e.g., Burger v. Kemp, 483 U.S. 776, 789-92, 107 S.Ct. 3114, 97 L.Ed.2d 638 (1987), this is no longer the case. In Williams v. Taylor, 529 U.S. 362, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000), Wiggins v. Smith, 539 U.S. 510, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003), and Rompilla v. Beard, 545 U.S. 374, 125 S.Ct. 2456, 162 L.Ed.2d 360 (2005), counsel conducted some inquiries, but the Court required a more robust, complete investigation, tethered at minimum to the norms of adequate investigation articulated by the American Bar Association Standards for Criminal Justice. Because of counsels’ deficient investigations, the Supreme Court overturned the petitioners’ death sentences in each case. In Williams v. Taylor, 529 U.S. 362, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000), the Supreme Court, for the first time under Strickland’s two-part test, reversed a death sentence based on an ineffective assistance of counsel claim. See Jenny Roberts, Too Little, Too Late: Ineffective Assistance of Counsel, The Duty to Investigate, and Pretrial Discovery in Criminal Cases, 31 Fordham Urb. L.J. 1097, 1110 (2004). The Court focused not on what was presented at trial, but on the adequacy of trial counsel’s preparation for the mitigation phase. Williams, 529 U.S. at 396, 120 S.Ct. 1495. To evaluate the reasonableness of trial counsel’s preparation, the Court looked to the standards for adequate investigation expressed in the ABA Standards for Criminal Justice. Id. at 396, 120 S.Ct. 1495 (citing 1 ABA Standards for Criminal Justice 4-4.1, cmt., p. 4-55 (2d ed.1980)). The Court found that undiscovered evidence of an abusive childhood and borderline mental retardation may have “influenced the jury’s appraisal of his moral culpability.” Id. at 398, 120 S.Ct. 1495. Soon after, the Court decided Wiggins v. Smith, 539 U.S. 510, 123 S.Ct. 2527, 156 L.Ed.2d 471 (2003), again focusing on trial counsel’s investigation and stressing that counsel, in most cases, must pursue all reasonable leads. In Wiggins, trial counsel employed a psychologist, used Social Services Records, and read through the Presentence Investigation Report to prepare for the mitigation phase. The Court nonetheless found counsel’s investigation unreasonable, as he did not prepare a forensic social history report as recommended by the ABA, and failed to pursue leads he already had suggesting his client suffered from a history of abuse and neglect. Id. at 524, 123 S.Ct. 2527. The Court stressed that it is “not only the quantum of evidence already known to counsel, but also whether the known evidence would lead a reasonable attorney to investigate further” that is the appropriate inquiry under Strickland. Id. at 527, 123 S.Ct. 2527. It also reminded us that physical and sexual abuse and diminished mental capacities compose the kind of “troubled history” that may diminish moral culpability. Id. at 535, 123 S.Ct. 2527. Most recently, in Rompilla v. Beard, 545 U.S. 374, 125 S.Ct. 2456, 162 L.Ed.2d 360 (2005), the Court reaffirmed that counsel’s unreasonably limited investigation cannot withstand scrutiny under Strickland. Counsel in Rompilla conducted a more thorough investigation than in the prior two cases, speaking with five family members and employing three experts who examined the defendant’s mental health at the time of the offense. Id. at 381-82, 125 S.Ct. 2456. Rompilla himself was unhelpful and even sent counsel on false leads. Id. at 381, 125 S.Ct. 2456. Despite this, the Court found counsel ineffective. The Court again relied on the ABA Standards, holding that counsel must investigate everything relevant to the penalty phase, regardless of the accused’s admissions or statements. Rompilla, 545 U.S. at 387, 125 S.Ct. 2456 (quoting 1 ABA Standards for Criminal Justice 4-4.1 (2d. ed 1982 Supp.)). As in Wiggins, knowledge of potential leads was critical in triggering the duty to continue investigating, and as in both prior cases, the Court found that the evidence of child abuse and mental health problems would have been persuasive in the mitigation phase. Id. at 392-93, 125 S.Ct. 2456. These cases stand for three important principles. First, the question is not whether counsel did something-, counsel must conduct a full investigation and pursue reasonable leads when they become evident. See Dickerson v. Bagley, 453 F.3d 690, 693 (6th Cir.2006) (stating that in Williams and Wiggins, the Supreme Court “made it clear and c[a]me down hard on the point that a thorough and complete mitigation investigation is absolutely necessary in capital cases.”); Smith v. Dretke, 422 F.3d 269, 278-79 (5th Cir.2005) (same). Second, to determine what is reasonable investigation, courts must look first to the ABA guidelines, which serve as reference points for what is acceptable preparation for the mitigation phase of a capital case. Rompilla, 545 U.S. at 387 n. 7, 125 S.Ct. 2456 (“[The 1989] Guidelines appl[y] the clear requirements for investigation.... ”); Wiggins, 539 U.S. at 524, 123 S.Ct. 2527 (“[T]he standards for capital defense work articulated by the American Bar Association (ABA) ... [should be used] as ‘guides to determining what is reasonable.’ ”). Finally, because of the crucial mitigating role that evidence of a poor upbringing or mental health problems can have in the sentencing phase, defense counsel must pursue this avenue of investigation with due diligence. Our own Circuit has emphasized this guiding principle. In Smith v. Mullin, 379 F.3d 919, 942 (10th Cir.2004), we held that it was “patently unreasonable” for trial counsel to fail to present evidence of Smith’s borderline mental retardation, brain damage, and troubled childhood, and stated that this type of mitigating evidence “is exactly the sort of evidence that garners the most sympathy from jurors.” Though the state’s aggravating case was particularly strong, we reversed Smith’s death sentence because of the power the mitigating evidence had to explain his behavior to the jury. Id. at 944. As in Williams, Wiggins, and Rompilla, Petitioner’s complaint is based on his counsel’s limited investigation into potential mitigating evidence about Mr. Wilson’s mental health, as well as counsel’s failure to present the available mental health diagnoses. Specifically, Petitioner raises two objections to counsel’s pre-trial preparation: his failure to engage an expert until shortly before trial, and his failure to supply the expert with readily available relevant information. Petitioner raises one objection to counsel’s performance at trial: his failure to present the expert’s actual diagnoses to the jury. 1. Pre-Trial Preparation Deficiencies a. Delay in Hiring Mental Health Expert First, counsel hired Dr. Reynolds only three weeks prior to trial and met with him only two days before he testified, despite the fact that counsel was assigned to this case two years in advance of trial. Under the American Bar Association Guidelines, “preparation for the sentencing phase, in the form of investigation, should begin immediately upon counsel’s entry into the case.” ABA Guidelines 11.8.3 (1989). The reason for the ABA’s direction is obvious — there must be sufficient time for interviews, research, and adequate testing before strategic planning can even begin. Additionally, if counsel waits until immediately before trial, it is too late to correct any invalid tests or to pursue leads discovered during the testing process, a requirement for counsel to be effective. See Rompilla, 545 U.S. at 392, 125 S.Ct. 2456 (effective counsel would have conducted further investigation after discovering “red flags” in initial investigation). The rush to prepare will invariably lead to unnoticed and untapped resources. Mr. Wilson’s case exemplifies the problems with delaying the investigation. Dr. Reynolds did not have time to conduct additional testing to confirm a diagnosis of schizophrenia, nor could the defense team gather collateral evidence that might provide insight into Mr. Wilson’s psychology. In his affidavit, Dr. Reynolds states flatly that “I needed additional testing, and further collateral data to support [a schizophrenia] diagnosis” but “[unfortunately, there wasn’t enough time.” Pet. Addendum 2, at ¶ 4. He further stated that, with the additional family information provided by appellate counsel after conviction and the additional testing he could then perform, he was able to “reach a more .accurate diagnosis,” id. at ¶ 7, and that this would have improved his testimony “enormously” and “helped the jury better understand Micheal’s [sic] emotional illness and how he could have participated in the crime.” Id. at ¶ 15. He stated his opinion that among “the most important and significant data to tell the jury was Mr. Wilson’s diagnosis of paranoid personality disorder,” or schizophrenia. Id. at ¶ 14; see also id. at ¶ 8 (“Mr. Wilson meets the criterion for a diagnosis of schizophrenia, paranoid type.... ”). Time constraints due to counsel’s tardy preparation precluded this more accurate and helpful diagnosis and testimony. Id. at ¶ 4. In Anderson v. Simmons, 476 F.3d 1131 (10th Cir.2007), we concluded that trial counsel in a capital case was constitutionally ineffective, partly on the ground that the investigator assigned to investigate the case in mitigation did not begin his work until the month before trial. Id. at 1143. Citing the 2003 version of the ABA Guidelines 10.7 Commentary, the Court regarded this delay as indicative of ineffectiveness, even without specific evidence regarding the consequences of delay. Id. (“The mitigation investigation should begin as quickly as possible, because it may affect the investigation of first phase defenses ..., decisions about the need for expert evaluations ..., motion practice, and plea negotiations.”). We regard the timeliness of engagement of expert evaluators as no less important. The Ninth Circuit reversed a denial of habeas corpus on grounds of ineffective assistance in a case with similar facts. In Bloom v. Calderon, 132 F.3d 1267, 1271 (9th Cir.1997), defense counsel contacted a psychiatric expert twenty days before trial — a period almost identical to the three weeks in this case. “Because counsel did not acquire the services of this key witness until days before trial, a hurried and inaccurate report resulted.” Id. at 1277. The court roundly rejected the claim that delay of this sort could be regarded as “strategic,” commenting that this would “ ‘strip[ ] that term of all substance.’ ” Id. (quoting Sanders v. Ratelle, 21 F.3d 1446, 1456 (9th Cir.1994)). The court concluded: “The complete lack of effort by Bloom’s trial counsel to obtain a psychiatric expert until days before trial, combined with counsel’s failure to adequately prepare his expert and then present him as a trial witness, was constitutionally deficient performance.” Id. Below, we explain why this case, like Bloom, involved inadequate preparation and presentation of the expert witness, as well as delay in engagement. We are, therefore, driven to the same conclusion. Judge Tymkovieh argues, in dissent, that the decision not to readminister the MMPI-2 test or collect collateral data was attributable not to the lack of time, but to Dr. Reynolds’ failure to recommend further testing to counsel: “Wilson’s counsel appropriately relied on Dr. Reynolds to decide how many rounds of mental health testing should be conducted. The record does not reveal that Dr. Reynolds ever advised counsel further testing beyond the initial round was necessary or advisable.” Diss. Op. 1133-34; see also id. at 1144 (“counsel did not consider additional tests and interviews necessary”). We do not read the record that way. According to his affidavit, Dr. Reynolds administered the MMPI-2 test during the first battery of testing, but (for an unexplained reason) the results were invalid. “[Bjecause his first MMPI-2 was invalid,” Dr. Reynolds explains, “I needed additional testing, and further collateral data to support this diagnosis. Unfortunately, there wasn’t enough time to obtain this information before trial.” Pet. Addendum 2, at ¶ 4. The failure to retest had nothing to do with whether Dr. Reynolds “advised counsel further testing beyond the initial round was necessary or advisable.” Dr. Reynolds did not need authorization from counsel to redo the invalid test. According to his affidavit, his inability to administer the MMPI-2 a second time was due to the lack of time. That was counsel’s fault for not engaging Dr. Reynolds until just before trial. The dissent also suggests that no negative consequences flowed from the late hiring because “[i]f counsel had considered further investigation necessary, he could have sought a continuance or conducted what further investigation was possible in the time remaining.” Diss. Op. 1144 (emphasis omitted). Perhaps so, but that does not make his performance any less ineffective. It might well be regarded as exacerbating counsel’s ineffective performance that he did not take any steps to repair the damage of his late start even when the consequences became apparent. Dr. Reynolds, in his affidavit, stated as follows: “Evaluations were performed on 1-22-97, 1-29-97, and 2-06-97. Results were made available to defense counsel shortly after February 7th, 1997.” Pet. Addendum 2, at ¶ 13. The MMPI-2 test “provided some evidence for a diagnosis of schizophrenia,” id. at ¶ 4, but was invalid and had to be administered again. If counsel did not grasp the importance of obtaining an accurate diagnosis and deliberately chose to do nothing, as the dissent seems to suggest, this is but confirmation of his lack of understanding of the role of mental health evidence in the mitigation phase of a capital trial. In any event, it is far from clear that if counsel had requested a continuance, the judge would have granted the motion. The trial court might well have been reluct